{"id":"W4406892577","doi":"10.1109/fllm63129.2024.10852425","title":"Screening Automation for Systematic Reviews: A 5-Tier Prompting Approach Meeting Cochrane’s Sensitivity Requirement","year":2024,"lang":"en","type":"article","venue":"","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Public Health Agency of Canada","funders":"","keywords":"Sensitivity (control systems); Automation; Computer science; Tier 1 network; Systematic review; Engineering; World Wide Web; MEDLINE; Electronic engineering; The Internet","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14069,0.002979555,0.00442547,0.01406683,0.002377331,0.007901189,0.003595273,0.003616907,0.01040338],"category_scores_gemma":[0.3933343,0.003112915,0.005076397,0.008069667,0.001612551,0.008176615,0.008450976,0.003095472,0.004519417],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002774094,"about_ca_system_score_gemma":0.01656933,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002935309,"about_ca_topic_score_gemma":0.007958459,"domain_scores_codex":[0.85066,0.09746598,0.02633812,0.008312072,0.01578851,0.001435368],"domain_scores_gemma":[0.4824082,0.4116873,0.02452777,0.03467236,0.04351753,0.00318685],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002763125,0.0004795244,0.008941851,0.02441358,0.001292536,0.0007580289,0.005469026,0.01135786,0.02474125,0.01220755,0.03429508,0.8732806],"study_design_scores_gemma":[0.00384589,0.003033131,0.02496886,0.01295403,0.005079114,0.00476812,0.003190033,0.4294302,0.07495172,0.2044905,0.2316083,0.001680094],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.009497172,0.001772731,0.953545,0.003532461,0.0003234207,0.005913835,0.002158132,0.02174252,0.001514733],"genre_scores_gemma":[0.03204022,0.0003735461,0.9636368,0.0007872301,0.0001007083,0.001611867,0.0007059943,0.0004160453,0.0003275293],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.85931,"threshold_uncertainty_score":0.744049,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7465126771264413,"score_gpt":0.5402911369114785,"score_spread":0.2062215402149629,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}